Race condition in cache invalidation when multiple workers update same key

[Katomoto]

Under concurrent load, multiple workers occasionally read a stale cache entry after invalidation because the delete and subsequent write are not atomic.

The issue appears when:

• Worker A reads key → cache miss → fetches from DB
• Worker B invalidates the key
• Worker A writes stale data back to cache
• Subsequent reads get the stale value until next TTL expiry

This is a classic cache stampede / stale write race. Fix options:

1. Use compare-and-swap (CAS) on write
2. Add a short lock per cache key during refresh
3. Adopt a read-through pattern so only one worker refreshes at a time.